[asterisk-dev] [Code Review] Call Completion: Asterisk Component

Mark Michelson mmichelson at digium.com
Thu Feb 25 10:32:09 CST 2010 


> On 2010-02-24 18:27:56, rmudgett wrote:
> > /trunk/main/ccss.c, lines 2106-2108
> > <https://reviewboard.asterisk.org/r/523/diff/3/?file=8306#file8306line2106>
> >
> >     Is it safe to reference interfaces after the channel is unlocked?

Same goes here as it does for my comment above.


> On 2010-02-24 18:27:56, rmudgett wrote:
> > /trunk/main/ccss.c, lines 2008-2009
> > <https://reviewboard.asterisk.org/r/523/diff/3/?file=8306#file8306line2008>
> >
> >     Is it safe to reference cc_interfaces after the channel unlock?
> >     
> >     The only cc_interface values used after the unlock is interface_tree and core_id.

After thinking about things more, I really haven't been taking into account the fact that channels are refcounted. If I have a pointer to a channel, then I have a reference to it, meaning that the channel cannot be freed out from under me. As such, by having a reference to the channel, the datastore cannot be freed out from under me either. What I have to be worried about more is another thread attempting to change the datastore's data.

So to answer your question, it's safe in the sense that a crash or other catastrophic failure will not happen. Of course, when dealing with data that can be changed by another thread, we either need to save off its value prior to unlocking the channel or only access it while the channel is locked. Of course, in this particular case, we're dealing with data that is set once and not changed by other threads anyway.

One thing I'm starting to realize after typing this up is that I really don't need to ref and unref the interface_tree in this function. As long as I have a channel reference, I don't have to worry about the interface_tree's refcount. I'll be making some changes with this in mind.


- Mark


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/523/#review1592
-----------------------------------------------------------


On 2010-02-24 17:00:54, Mark Michelson wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/523/
> -----------------------------------------------------------
> 
> (Updated 2010-02-24 17:00:54)
> 
> 
> Review request for Asterisk Developers.
> 
> 
> Summary
> -------
> 
> This is it folks. CCSS is at a point where a review will be beneficial.
> 
> CCSS stands for Call Completion Supplementary Services. An admittedly out-of-date
> overview of the architecture can be found in the file doc/CCSS_architecture.pdf
> in the CCSS branch. Off the top of my head, the big differences between what is
> implemented and what is in the document are as follows:
> 
> 1. We did not end up modifying the Hangup application at all.
> 2. The document states that a single call completion monitor may be used across
>    multiple calls to the same device. This proved to not be such a good idea
>    when implementing protocol-specific monitors, and so we ended up using one
>    monitor per-device per-call.
> 3. There are some configuration options which were conceived after the document
>    was written. These are documented in the ccss.conf.sample that is on this
>    review request.
>    
> For some basic understanding of terminology used throughout this code, see the
> ccss.tex document that is on this review.
> 
> This implements CCBS and CCNR in several flavors.
> 
> First up is a "generic" implementation, which can work over any channel technology
> provided that the channel technology can accurately report device state. Call
> completion is requested using the dialplan application CallCompletionRequest and can
> be canceled using CallCompletionCancel. Device state subscriptions are used in order
> to monitor the state of called parties.
> 
> Next, there is a SIP-specific implementation of call completion. This method uses the
> methods outlined in draft-ietf-bliss-call-completion-06 to implement call completion
> using SIP signaling. There are a few things to note here:
> 
> * The agent/monitor terminology used throughout Asterisk sometimes is the reverse of
>   what is defined in the referenced draft.
> 
> * Implementation of the draft required support for SIP PUBLISH. I attempted to write
>   this in a generic-enough fashion such that if someone were to want to write PUBLISH
>   support for other event packages, such as dialog-state or presence, most of the effort
>   would be in writing callbacks specific to the event package.
> 
> * A subportion of supporting PUBLISH reception was that we had to implement a PIDF
>   parser. The PIDF support added is a bit minimal. I first wrote a validation
>   routine to ensure that the PIDF document is formatted properly. The rest of the
>   PIDF reading is done in-line in the call-completion-specific PUBLISH-handling
>   code. In other words, while there is PIDF support here, it is not in any state
>   where it could easily be applied to other event packages as is.
> 
> Finally, there are a variety of ISDN-related call completion protocols supported. These
> were written by Richard Mudgett, and as such I can't really say much about their
> implementation. I'm leaving it to Richard to add any comments he wants about this
> matter. The libpri component of call completion support is posted as review 522
> on ReviewBoard.
> 
> The code added here is somewhat massive, so questions are welcome as much as critiques.
> Happy reviewing!
> 
> 
> Diffs
> -----
> 
>   /trunk/CHANGES 248658 
>   /trunk/apps/app_dial.c 248658 
>   /trunk/channels/chan_dahdi.c 248658 
>   /trunk/channels/chan_local.c 248658 
>   /trunk/channels/chan_sip.c 248658 
>   /trunk/channels/sig_analog.h 248658 
>   /trunk/channels/sig_analog.c 248658 
>   /trunk/channels/sig_pri.h 248658 
>   /trunk/channels/sig_pri.c 248658 
>   /trunk/channels/sip/include/sip.h 248658 
>   /trunk/configs/ccss.conf.sample PRE-CREATION 
>   /trunk/configs/chan_dahdi.conf.sample 248658 
>   /trunk/configure.ac 248658 
>   /trunk/doc/tex/asterisk.tex 248658 
>   /trunk/doc/tex/ccss.tex PRE-CREATION 
>   /trunk/funcs/func_callcompletion.c PRE-CREATION 
>   /trunk/include/asterisk/ccss.h PRE-CREATION 
>   /trunk/include/asterisk/channel.h 248658 
>   /trunk/include/asterisk/channelstate.h PRE-CREATION 
>   /trunk/include/asterisk/devicestate.h 248658 
>   /trunk/include/asterisk/frame.h 248658 
>   /trunk/include/asterisk/manager.h 248658 
>   /trunk/include/asterisk/rtp_engine.h 248658 
>   /trunk/include/asterisk/xml.h 248658 
>   /trunk/main/acl.c 248658 
>   /trunk/main/asterisk.c 248658 
>   /trunk/main/ccss.c PRE-CREATION 
>   /trunk/main/channel.c 248658 
>   /trunk/main/xml.c 248658 
> 
> Diff: https://reviewboard.asterisk.org/r/523/diff
> 
> 
> Testing
> -------
> 
> Too much to list :)
> 
> For a decent look at the tests executed, see the "Testing Done" section of review request 410.
> Those tests have all been run using combinations of generic, SIP, and ISDN agents and monitors.
> 
> 
> Thanks,
> 
> Mark
> 
>

More information about the asterisk-dev mailing list